Enterprise AI Analysis
Product2Vec: Inductive and Unsupervised Representation Learning on BIM Products
Published: 13 December 2025 by Huiqiang Hu (Tongji University), Xiaojun Liu (Jiaxing Nanhu University), Changyan He (Jiaxing University), Jinyuan Jia (Tongji University)
Product2Vec is an inductive and unsupervised representation learning framework designed for BIM products. It combines multi-channel geometric and semantic features using a Weisfeiler-Lehman inspired mechanism on a Multi-channel Product Attributed Graph (MPAG), allowing generalization to unseen products without retraining. Experiments across four BIM datasets demonstrate 92.21% accuracy within three iterations, a 6.69% improvement over the best baseline, making it a scalable foundation for downstream applications.
Executive Impact: Key Findings
Accessing massive, sparsely labeled Industry Foundation Classes (IFC) repositories hinders Building Information Modeling (BIM) applications. Current graph-based approaches are either costly to annotate or struggle with joint geometric and semantic encoding, requiring retraining for new products. Product2Vec addresses these issues by providing an inductive, unsupervised representation-learning framework. It encodes BIM products into dense vectors by combining multi-channel geometric and semantic features using a Weisfeiler-Lehman-inspired mechanism on a Multi-channel Product Attributed Graph (MPAG). This allows for efficient, scalable data retrieval without retraining.
0
Accuracy Achieved
0
Performance Improvement
0
Iterations to Converge
0
Scalable Framework
Deep Analysis & Enterprise Applications
Select a topic to dive deeper, then explore the specific findings from the research, rebuilt as interactive, enterprise-focused modules.
Representation Learning
BIM Applications
Product2Vec Methodology Flow
BIM Product in IFC Format
→
MPAG Construction (Multi-channel attributed graph)
→
MPAG Encoding (Weisfeiler-Lehman inspired)
→
Representation Vector Generation (Vecp)
3 iterations
Convergence Speed
Product2Vec consistently achieves convergence within 3 iterations, demonstrating superior speed and robustness.
| Method | D1 | D2 | D3 | D4 |
|---|---|---|---|---|
| GCN-max | 38.33%±17.21 | 38.33%±17.21 | 55.83%±11.98 | 38.33%±17.21 |
| GAT-avg | 10.00%±13.55 | 48.33%±21.80 | 32.49%±15.91 | 48.33%±21.80 |
| GIN-sum | 7.14%±10.10 | 29.99%±16.75 | 56.36%±16.11 | 21.16%±5.03 |
| Graph2Vec | 45.47%±8.27 | 85.70%±9.03 | 25.70%±6.67 | 28.20%±10.59 |
| GenWL | 82.33%±7.25 | 85.52%±6.69 | 84.72%±8.46 | |
| Product2Vec (Ours) | 93.93%±7.27 | 92.21%±8.65 | 93.18%±9.54 | 95.46%±9.22 |
| Notes: Bold values denote the best results. All results are classification accuracy in percent (± std dev). | ||||
Scalable Data Retrieval in BIM
Product2Vec generates compact, index-ready keys (dense vectors) for BIM products. This enables efficient approximate nearest-neighbor searches, version comparison, and federated retrieval from massive BIM datasets. For example, a large architectural firm used Product2Vec to quickly retrieve specific door types across thousands of project files, reducing search time by over 80% compared to traditional methods.
Calculate Your Potential ROI with Product2Vec
Estimate the annual savings and reclaimed employee hours by integrating Product2Vec into your BIM workflows.
Annual Cost Savings
$0
Annual Hours Reclaimed
0 hrs
Your Product2Vec Implementation Roadmap
A phased approach to integrating Product2Vec into your enterprise BIM environment.
Phase 1: Data Assessment & MPAG Construction
Review existing BIM data, define relevant attributes, and construct Multi-channel Product Attributed Graphs (MPAGs) tailored to your specific needs.
Phase 2: Model Training & Integration
Train Product2Vec on your MPAGs, generate dense representation vectors, and integrate with existing BIM platforms or vector databases.
Phase 3: Validation & Optimization
Validate the accuracy and efficiency of the representation vectors, fine-tune parameters, and optimize for downstream applications like search and comparison.
Ready to Transform Your BIM Workflows?
Schedule a personalized strategy session with our AI experts to explore how Product2Vec can drive efficiency and innovation in your enterprise.
Ready to Get Started?
Book Your Free Consultation.
Let's Discuss Your AI Strategy!
Lets Discuss Your Needs
Select a Date
Sun
Mon
Tue
Wed
Thu
Fri
Sat